An introduction to epidemiological terminology
Author: Mira Katan Kahles, Department of Neurology, University Hospital of Zurich, Switzerland
When reading clinical research papers often the words “controlling for confounding” or “adjusting for potential confounders” are used, but what exactly is the meaning of these sentences?
In clinical stroke investigations, especially in the setting of case control or cohort studies researchers try to find “true” associations of one variable of interest with another variable. However, a factor may seem to be linked with another factor although in truth there is no such independent association. This can happen when there is “confounding” at work. Confounding occurs- according to Fletcher and Fletcher 1 –if two factors are associated with each other but only one of the two factors is in truth association with the outcome of interest1.
For example the hypothesis is that “grey hair “ is a risk factor for stroke mortality. Now when assessing the association of the occurrence of grey hair with mortality after stroke we might find indeed an association. However when “adjusting” for age, which is also, associated with stroke mortality the association of grey hair with stroke could go away. This suggests that the relationship between “grey hair “ and stroke mortality is confounded by age, which is related to both “grey hair” and stroke mortality.
Thus some key vascular risk factors such as age, smoking, diabetes, hypertension among others, are almost always adjusted for because stroke outcomes including mortality vary according to them.
The potential for confounding does not mean that it is actually present or, if present in a particular study, has a big enough effect on the results to matter. Nevertheless, one should know where and how to look for it 1. For example, think about potential confounders before you plan a study and accordingly plan to collect all the necessary data in order to be able to adjust for these potential confounders.
Finally, there are always potential confounding factors, which cannot be adjusted for because we do not know them, thus a real causal relationship of a variable with another in clinical research can only be assessed in a randomized controlled trial setting.
Reference source: 1. Clinical Epidemiology –THE ESSENTIAL by R. H. Fletcher, S. W. Fletcher and G.S. Fletcher